Anda di halaman 1dari 8

Precision Engineering 28 (2004) 232239

Short communication

Uncertainty analysis of positional deviations of CNC machine tools


Ignacio Lira , George Cargill
Departamento de Ingenier a Mecnica y Metalrgica, Ponticia Universidad Catlica de Chile, Vicua Mackenna, 4860 Santiago, Chile Received 13 December 2002; accepted 30 June 2003

Abstract The assessment of the measurement uncertainty is an indispensable task in all calibration procedures. By international accord, the evaluation is to be done in accordance with the ISO Guide to the Expression of Uncertainty in Measurement (GUM). To calibrate the positional deviations of computer numerically controlled (CNC) machine tools, calibration laboratories will usually follow the guidelines in ISO 230-2 International Standard. However, that standard does not address uncertainty. In this paper, we present an uncertainty evaluation scheme that is rmly grounded in the GUM, and can therefore be of use as a guide to develop appropriate uncertainty calculations in this and similar types of calibrations. 2003 Elsevier Inc. All rights reserved.
Keywords: Uncertainty analysis; Machine tools; Calibration; Dimensional metrology

1. Introduction The ISO 230-2 International Standard [1] is accepted and used worldwide to determine the accuracy and repeatability of positioning of computer numerically controlled (CNC) machine tool axes; it may be used for type testing, acceptance tests, comparison testing, periodic verication and machine compensation. The test method standardized in ISO 230-2 involves repeated measurements at predened positions of the axis under test. The results are used to determine various parameters intended to quantify the performance of the machine. Some of these parameters depend only on the values measured at a given position; other parameters characterize the global behavior of the axis. The local parameters are the mean positional deviations, the reversal values and the repeatabilities. The global parameters are the reversal value, the mean reversal value, the repeatability, the systematic positional deviation, the mean positional deviation and the accuracy. The repeatability and accuracy depend upon a further local parameter that ISO 230-2 denes as the standard uncertainty of positioning, s. For example, the unidirectional repeatability at position i is dened as 4si . The intention is clearly to provide a range derived from an expanded uncertainty using a coverage factor of 2. The latter terminology is taken from the ISO Guide to the Expression of Uncertainty in Measure Corresponding

author. Tel.: +56-2-686-4629; fax: +56-2-686-5828. E-mail address: ilira@ing.puc.cl (I. Lira).

ment (GUM) [2], which ISO 230-2 declares to follow. However, the GUM denes the parameter s as the sample standard deviation; under no circumstances it should be taken as an estimator of standard uncertainty. It follows that there is a conict in terminology between these two documents. Moreover, the International Vocabulary of Basic and General Terms in Metrology (VIM) [3] denes calibration as the set of operations that establish the relationship between values of quantities indicated by a measuring system and the corresponding values realized by measurement standards. Since the test code specied by ISO 230-2 clearly complies with this (abridged) denition, it follows that the test report can be interpreted as a calibration certicate of the machine. In that case, however, the uncertainty of measurement should be reported together with the measurement results [4]. ISO 230-2 does not address this crucial point. The purpose of this paper is, therefore, to propose a procedure by which to evaluate the uncertainty associated with the measured positional deviations of machine tools. The procedure follows strictly the recommendations of the GUM and uses the measurement results obtained with the ISO 230-2 test method. We propose that future revisions of the standard should include this procedure. In this way, the testing laboratory would be able to issue a calibration certicate instead of just a test report. Future revisions of ISO 230-2 should also alter slightly its terminology with regard to the parameter s, so as to comply with the GUM. The outline of the paper is as follows. In Section 2 we summarize the ISO 230-2 test method. In Section 3 we present

0141-6359/$ see front matter 2003 Elsevier Inc. All rights reserved. doi:10.1016/j.precisioneng.2003.06.001

I. Lira, G. Cargill / Precision Engineering 28 (2004) 232239

233

the measurement model on which we base the uncertainty evaluation. From this model all relevant input quantities are identied. The standard uncertainties of these quantities are derived following the recommendations of the GUM. An example is presented in Section 4 and conclusions are given in Section 5. Appendix A treats the case of nonlinear measurement models.

Table 1 Local parameters dened in ISO 230-2 Parameter Mean positional deviations Symbol x i x i x i Reversal value Bi
2 si

Denition (1/n)
j

xij

(1/n) j xij (x i + xi )/2 x i xi


j (xij j (xij

2. The ISO 230-2 test method In the ISO 230-2 test method the machine is programmed to locate its moving part at a series of predened target positions along the axis under test. Fig. 1 shows the standard test cycle. It consists of n unidirectional approaches in both the positive (increasing) and negative (decreasing) directions to m target positions. The symbols and are used to represent each of these directions of movement. The measured quantities are the positional deviations. For point i in unidirectional approach j, these are dened as the differences between the actual positions pij and the target positions pi : xij | = pij | pi , (1)

Experimental variancesa Repeatabilities

xi )2 /(n 1) xi )2 /(n 1)

2 si

Ri Ri Ri

4si 4si Max[2si +2si +|Bi |; Ri ; Ri ]

a ISO 230-2 does not actually dene the experimental variances. It denes their square roots as estimators of the unidirectional standard uncertainty of positioning.

where the symbol | stands for either or . In this expression the index j runs from 1 to n and the index i runs from 1 to m. In the standard test cycle, n = 5. The value of m depends on the length of the axis; ISO 230-2 recommends selecting a minimum of ve unevenly spaced target positions per meter of measurement travel. Local and global parameters are derived from the positional deviations; they are given in Tables 1 and 2, respectively. It is recommended that these parameters be shown in
Position i (m = 8) 1 2 3 4 5 6 7 8

graphical form. Fig. 2 shows a typical plot; it is taken from the example provided in ISO 230-2. The test report should also provide information regarding the test equipment, the test conditions, the built-in compensation routines used (if any), the feed rate, the dwell time at the target positions, the warm-up operations that precede the test, etc. In particular, the test report should detail the location of the temperature sensors, the indicated temperatures and the thermal expansion coefcient used to calculate the nominal differential expansion (NDE) correction for measurements performed at temperatures other than 20 C. Although the standard provides a typical value of uncertainty equal to 2 m/(m C) resulting from temperature departures from 20 C, this is left for the user of the test report to assess.
Table 2 Global parameters dened in ISO 230-2 Parameter Reversal value Symbol B B R R R E E E Mean positional deviation M Aa A A Denition Max[|Bi |] (1/m)
i

Mean reversal value Repeatabilities

Bi

Max[Ri ] Max[Ri ] Max[Ri ] Max[x i ] min[x i ] Max[x i ] min[x i ] Max [x i ; x i ] min [x i ; x i ] Max[x i ] min[x i] Max[x i +2si ] min[x i 2si ] Max[x i +2si ] min[x i 2si ] i +2si ] Max[x i +2si ; x i 2si ] min[x i 2si ; x

3 Cycle j (n = 5) 4

Systematic positional deviations

Accuracies

= target positions
Fig. 1. Standard test cycle for n = 5 bidirectional approaches to m = 8 calibrated points.

a In ISO 230-2 the parameter A is given as max[x i +2si ]min[x i +2si ].

234

I. Lira, G. Cargill / Precision Engineering 28 (2004) 232239

6 4 2 Deviation (m)

xi 2 si or xi 2 si

R1
0

B1

R
-2 -4 -6 -8 0 500 1000 1500

xi xi xi

2000

Posi tion (mm)


Fig. 2. Example of results obtained with the test code in ISO 230-2.

It is then clear that the main contribution of ISO 230-2 is the standardization of the various parameters that characterize the behavior of machine tools, such as repeatability and accuracy. However, it does not address the calculation of the measurement uncertainty. It does not consider resolution, misalignment, and calibration of the measuring equipment. In addition, as mentioned above, its terminology does not comply with that in the GUM.

eralization of this law to more than two input quantities is straightforward. 3.1. Basic model In the calibration of machine tools, the measurands are the positional deviations. These are modeled as xi | = pi | pi , (4)

3. A proposal to evaluate the uncertainty in the test of CNC machines Reference [4] indicates that the statement of the result of a measurement is complete only if it contains both the value attributed to the measurand and the uncertainty of measurement associated with that value. In order to evaluate measurement uncertainties, the measurands must be dened through appropriate measurement models. When the models are linear or weakly nonlinear, they allow to express the standard uncertainties of the output quantities in terms of the standard uncertainties of the input quantities by using the so-called law of propagation of uncertainties, LPU [2,5]. This law says that for a quantity modeled as z = f(x, y), the standard uncertainty is obtained as the square root of
2 2 2 2 u2 [z] = cx u [x] + cy u [y] + 2cx cy rxy u[x]u[y],

where pi | is the corrected indication of the measurement system at position i and pi is the corrected target position. As in Section 2, the symbol | stands for either or . The models for the quantities pi | and pi should in turn include all conceivable random and systematic effects that inuence them. While this depends on the particulars of the measurement system, it is reasonable to assume that measurements in testing machine tools are in general affected or limited by the characteristics of the test equipment and also by the conditions at which the test is carried out. The latter include temperature, alignment and resolution. Therefore, we write pi | = ft fa pmi | + cs and pi = pni + cr , where
j

(5)

(2)

(6)

(3)

pmi | =

pij |

(7)

where cx = f /x, cy = f /y and rxy is the correlation coefcient between the input quantities x and y. If these are independent, the correlation coefcient is zero. The gen-

is the arithmetic mean of the actual positions pij | in the n unidirectional approaches to the nominal (programmed) target position pni , ft is a correction factor to compensate for thermal

I. Lira, G. Cargill / Precision Engineering 28 (2004) 232239

235

expansion of the positioning scale, fa is a correction factor to compensate for possible misalignment of the measurement system, cs is a correction associated with the measurement system, and cr is a correction associated with the resolution of the axis scale. (Note the difference between our denition of xi | and the parameter x i | in ISO 230-2 standard. In terms of our notation, the latter is x i | = pmi | pni .) Replacing (5) and (6) in (4), using the LPU, assuming that all input quantities are independent, and recognizing that no uncertainty should be attached to the given nominal positions, we obtain u2 [xi |] = (ft fa u[pmi |])2 + (fa pmi |u[ft ])2 + (ft pmi |u[fa ])2 + u2 [cr ] + u2 [cs ]. (8) Finally, the model for the mean bidirectional positional deviation at position i is
1 x i = 2 (xi +xi ).

3.3. Type B uncertainties The type A evaluation of standard uncertainty applies only to quantities that are measured directly several times under repeatability conditions. The uncertainty of input quantities that are measured only once, that are evaluated from models that involve further quantities, or that are imported from other sources, should be evaluated by other means. The GUM refers to these other means as the type B method of evaluation. In many cases this involves obtaining an uncertainty as the standard deviation of the probability density function (pdf) that is assumed to apply to the quantity involved. Consider rst the NDE correction. The model for this correction is derived from the equation that denes the linear coefcient of thermal expansion . This equation is L Lo = Lo (T To ), (13)

(9)

The uncertainty associated with this quantity is then


1 2 u 2 [x i] = 4 (u [xi ] + u2 [xi ] + 2u[xi ]u[xi ]).

(10)

This last expression assumes a correlation coefcient between xi and xi equal to one. This is reasonable, since all inuence quantities that are likely to affect xi , such as barometric pressure, ambient temperature and humidity, are likely to affect xi in the same sense. 3.2. Type A uncertainties According to the GUM, the squares of the standard uncertainties associated with the unidirectional position means pmi | should be evaluated as the sample variances divided by the number n of unidirectional measurements. Thus, we set u2 [pmi |] = where Si |2 =
2 j (pij | pmi |)

where Lo is the length of an object at a reference temperature To and L is the length of the same object at the actual temperature T. In our case, Lo is the position poi at temperature To = 20 C and L is the measured position pmi | at the temperature of the test. Therefore, the thermal correction factor is ft = (1 + T)1 , where u2 [ft ] = ( T 2 u2 [] + 2 u2 [ T ])ft4 . (14) T = T 20 C. Application of the LPU then gives (15)

Si |2 , n .

(11)

The temperature T should be measured at at least two places along the scale and at various times, for example, at the beginning and at the end of the cycle. Doing this allows to establish a range of values for T. This range should be increased to both sides by including the expanded uncertainty of the temperature measurement device. In this way one obtains a range wT over which a uniform pdf is a reasonable assumption. Thus, from Equation 3.19 in Reference [5] we set u2 [ T ] = w2 T . 12 (16)

(n 1)

(12)

Note the difference between the sample variances Si |2 and si |2 (see the ISO 230-2 denition in Table 1). The former uses the actual positions pij | while the latter uses the deviations xij |. If the measuring instrument indicates directly the deviations, the actual positions must be obtained as pij | = xij | + pni . Eq. (11) is the well-known type A evaluation of standard uncertainty [2]. In this method of evaluation, the sample standard deviation Si has to be divided by the square root of the number of measurements n to obtain the standard uncertainty u[pmi |]. This is due to the fact that u[pmi |] is associated with the average of the measurements, not to any one single result pij |. For this reason we believe that the quantities si | in Table 1 should not be called estimators of standard uncertainties. For further discussion on this point, see Section 3.8 of Reference [5].

The coefcient is normally not measured; its value is imported from the manufacturer of the machine or from tabulated values in handbooks for the material one believes the scale to be made off. The standard uncertainty u[] should then be evaluated from a uniform pdf whose width w has to be assigned depending on how well one knows the material and on how well that materials coefcient is known. In other words, w corresponds to twice the maximum possible error in the assignment of the value of . With this assumption we get u2 [] = w2 . 12 (17)

It should be noted, however, that in the vicinity of T = 0 Eq. (15) yields a value for u[ft ] that is independent of the uncertainty in knowing . This is obviously an unreasonable result; it is due to the nonlinearity of the model for the thermal

236

I. Lira, G. Cargill / Precision Engineering 28 (2004) 232239

expansion correction factor in the vicinity of vanishing values of the input quantities. In the Appendix A we address this problem. (Note that by writing Eq. (13) in the form L Lo = L T , one gets ft = 1 T . An almost equal numerical value for the correction factor is obtained, but the expression for the standard uncertainty becomes u2 [ft ] = T 2 u2 [] + 2 u2 [ T ]. We feel that Eq. (15) is to be preferred. In any case, however, if T = 0 the uncertainty of ft is independent of the uncertainty of .) Consider next the alignment correction factor fa . This is modeled as fa = cos , (18) where is the angle between the actual measurement direction and the axis direction. Because of the form of (18) this correction factor is usually referred-to as the cosine error. Application of the LPU gives u[fa ] = sin . (19)

However, normally the best estimate for the misalignment angle will be = 0, yielding fa = 1 and u[fa ] = 0 irrespective of the uncertainty associated with . This is again an unreasonable result, and is due to the strong nonlinearity of cos in the vicinity of = 0. Therefore, we recommend to estimate the uncertainty of fa from
2 1 u 2 [f a ] = 2 (fa cos max + 1) fa ,

The value of csr is zero, and its uncertainty is obtained as 2 /12, where r is the resolution of the measureu2 [csr ] = rs s ment system. Since this resolution should normally be much smaller than r, the uncertainty u[csr ] can be neglected. The value and uncertainty of cc should be obtained from the calibration certicate of the instrument, thus providing the traceability for the calibration of the machine. Finally, if the instrument performs environmental corrections automatically, the value of the correction ce may be taken as zero. However, the uncertainty u[ce ] is normally not given in the calibration certicate of the measuring system. A reasonable assumption is to take this uncertainty as being proportional to the nominal position pni and to the temperature difference Te = Te To , where Te is the ambient temperature. In other words, we may write u[cei ] = Kpni Te , where the value for the proportionality constant K should be establish based on the characteristics of the measurement system. Although this procedure to estimate u[ce ] may be questionable, it is the most one can do if there is no information about the uncertainty of the environmental compensation. We must remember that uncertainties are parameters that are not measured, they only express the doubt that we have on the results of measurements [2]. Therefore, subjective criteria for establishing standard uncertainties do not contradict the GUM. 3.4. Expanded uncertainties To obtain the expanded uncertainties U[x i ], a coverage factor k = 2 is appropriate to dene intervals estimated to have a level of condence of about 0.95 for the true value of the mean bidirectional deviations x i . This is because the standard uncertainties u[x i ] are obtained through models that involve several input quantities whose standard uncertainties are derived mostly from type B analysis. Therefore, the number of degrees of freedom should be rather large (see Chapter 4 in Reference [5]).

(20)

where max is an upper bound for the maximum deviation angle, and where we must use sin max fa = . (21) max The angle max has to be established based on the particulars of the measurement system. Since normally this angle should be small, cos max 1 and u[fa ] 0. Eqs. (20) and (21) are derived in the Appendix A. In the Appendix A we also discuss why a uniform pdf is not appropriate to represent the cosine error. The evaluation of the resolution correction is straightforward. Its value is equal to zero and its standard uncertainty is obtained from a uniform pdf of width equal to the resolution r of the axis scale. Thus, u 2 [c r ] = r2 . (22) 12 The measurement system correction cs depends on the instrument that is used to perform the measurements. An appropriate model for this correction is cs = csr + cc + ce , where csr is a resolution correction, cc is a calibration correction and ce is a correction arising from use of the instrument at non-standard environmental conditions. From this model we get u2 [cs ] = u2 [csr ] + u2 [cc ] + u2 [ce ].

4. Example In this section, we calculate the uncertainty associated with the mean bidirectional deviations reported in the example given in ISO 230-2. The example refers to the test of a linear axis at m = 11 positions. Tables 3 and 4 show the unidirectional deviations xij and xij , respectively. Table 5 shows the numerical values of each uncertainty term in the right-hand side of (8) for the mean unidirectional positional deviations xi . The corresponding values for the decreasing deviations xi are very similar. The dominant contribution arises from the NDE correction. Table 6 shows the mean bidirectional positional deviations x i and their expanded uncertainties U[x i ]. These values are plotted in Fig. 3, and were obtained as follows. The test report given in ISO 230-2 mentions that NDE correction was performed, but it does not state the temperature

I. Lira, G. Cargill / Precision Engineering 28 (2004) 232239 Table 3 Deviations (in micrometers) for a series of 5 unidirectional approaches to 11 nominal positions in the increasing direction Position i 1 2 3 4 5 6 7 8 9 10 11 xi 1 1.2 0.5 0.2 0.6 1.9 3.0 3.7 3.7 3.5 3.2 3.6 xi 2 1.7 0.9 0.6 1.2 2.3 3.5 4.3 4.4 4.3 3.8 4.0 xi 3 1.9 1.1 0.7 1.3 2.9 3.7 4.6 5.1 5.0 4.7 4.5 xi 4 1.3 0.2 0.1 0.3 1.4 2.8 3.6 3.6 3.2 2.8 3.2 xi 5 1.9 0.8 0.7 1.3 2.3 3.7 4.5 4.6 4.5 4.1 4.5

237

Table 5 Components of the standard uncertainty (in m) associated with the positional deviations in the increasing direction Position i 1 2 3 4 5 6 7 8 9 10 11 ft fa u[pmi ] 0.148 0.158 0.201 0.206 0.248 0.186 0.206 0.282 0.330 0.334 0.254 fa pmi u[ft ] 0.030 0.772 1.561 2.319 3.106 3.890 4.658 5.445 6.213 6.971 7.724 ft pmi u[fa ] 0.003 0.079 0.161 0.239 0.320 0.400 0.479 0.560 0.640 0.718 0.795 u[cr ] 0.289 0.289 0.289 0.289 0.289 0.289 0.289 0.289 0.289 0.289 0.289 u[cs ] 0.003 0.019 0.038 0.056 0.075 0.094 0.113 0.132 0.150 0.169 0.187

The nominal positions are given in Table 6.

Table 4 Deviations (in m) for a series of 5 unidirectional approaches to 11 nominal positions in the decreasing direction Position i 1 2 3 4 5 6 7 8 9 10 11 xi 1 2.3 3.6 3.5 3.0 1.7 0.4 0.4 0.2 0.2 0.3 0.1 xi 2 2.1 3.5 3.3 2.7 1.5 0.2 0.7 0.6 0.2 0.1 0.6 xi 3 1.9 3.1 3.0 2.4 1.0 0.2 1.0 1.0 1.0 0.9 1.2 xi 4 2.8 3.7 3.8 3.2 1.9 0.9 0.0 0.2 0.5 0.5 0.4 xi 5 2.2 3.2 3.5 2.6 1.1 0.1 0.9 1.1 0.5 0.4 0.9

Table 6 Mean bidirectional positional deviation and expanded uncertainties (k = 2) pni (mm) 6.711 175.077 353.834 525.668 704.175 881.868 1055.890 1234.304 1408.462 1580.269 1750.920 x i (m) 0.3 1.4 1.5 0.9 0.4 1.6 2.4 2.4 2.1 1.9 2.2 U [x i ] (m) 0.7 1.7 3.2 4.7 6.3 7.9 9.4 11.0 12.5 14.0 15.6

The nominal positions are given in Table 6.

20 15 10 5 0 -5 -10 -15 -20 0 500 Position (mm)


Fig. 3. Mean bidirectional positional deviations and expanded uncertainties for the example of ISO 230-2.

Deviation (m)

x i + U( x i ) xi x i U( x i )

1000

1500

2000

238

I. Lira, G. Cargill / Precision Engineering 28 (2004) 232239

difference used. It indicates a minimum temperature of 21.8 C at the start of the test at position 50 mm, and a maximum temperature of 23.1 C at the end of the test at position 1700 mm. Therefore, we assume an average temperature T = 22.45 C and a temperature difference T = 2.45 C. To simulate the actual (uncorrected) positions, we take pij | = (xij | + pni )/ft with ft = 0.99997. For the uncertainty of T we take wT = (23.1 21.8) C = 1.3 C. Since this temperature spread is likely to be much greater than the uncertainty of the thermometer, we ignore the latter. This gives u[ T ] = 0.38 C. The thermal expansion coefcient is reported as being = 11 106 C1 . For the associated uncertainty we use w = 0.2, that is, we assume a maximum error of 10% of the nominal value to each side. This gives u[] = 0.635 106 C1 . For the maximum misalignment angle, we use max = 0.1 , giving u[fa ] = 0.45 106 . For consistency, in (8) we use the value fa = 0.99999949 obtained from (21). However, to evaluate the deviations xi and xi we take fa = 1 in order not to modify the measured values given in the example of ISO 230-2. From the nominal positions in the example we assume that the scale resolution is rs = 1 m. This gives u[cr ] = 0.289 m. The example does not give details on the measurement system used to obtain the data in Tables 3 and 4. We assume that a laser interferometer was used with rs = 0.01 m resolution, giving u[csr ] = 0.0029 m. We assume also that the calibration certicate of the interferometer states a negligible correction cc at standard conditions with a typical relative expanded uncertainty U [cc ] = 2 107 and a coverage factor k = 2. We then take u[cci ] = pni 107 , giving u[cc ] = 0.175 m at nominal position 1750.92 mm. Finally, the ambient temperature is reported as being 20.6 C at the start of the test and 20.9 C at the end. These temperatures provide only an indication of the environmental conditions during the test, but are not used for NDE correction. In our model we use ce = 0, Te = 20.75 C, Te = 0.75 C and K = 0.05 C1 . These values give a maximum uncertainty u[ce ] = 0.066 m at nominal position 1750.92 mm.

certainties together with types A and B evaluation methods. We propose to include this uncertainty evaluation procedure in a future revision of the standard. The revision should also alter slightly the terminology, so as to comply with the GUM.

Acknowledgments We acknowledge gratefully the suggestions offered by the referees to the manuscript submitted originally.

Appendix Both the thermal expansion and alignment correction factors arise frequently in dimensional measurements. Since the models for these quantities are nonlinear, one may encounter problems when applying the LPU to derive the corresponding uncertainties. Possible solutions are developed in this appendix. Consider rst the thermal expansion correction factor. For generality we write x instead of T, y instead of and z instead of ft , giving z= 1 . (1 + xy) (A.1)

Assume that x and y are independent and that the state of knowledge about these quantities is represented by pdfs fx (x) and fy (y), respectively. Use of the tools in Section 6.9 of Reference [5] gives the following pdf for z fz (z) = z2 where x = y1 (z1 1). (A.3) fx (x)fy (y)|y1 | dy, (A.2)

Assume now that x and y are distributed uniformly over the intervals (xm , xM ) and (ym , yM ), respectively. The integrand in (A.2) is then equal to the constant (wx wy )1 , where wx = (xM xm ) and wy = (yM ym ), when the following two inequalities are satised simultaneously xm x xM , (A.4) (A.5)

5. Conclusions ISO 230-2 standardizes the various parameters that characterize the behavior of machine tools, such as repeatability and accuracy. However, it does not address the calculation of uncertainty, which limits its use as reference for issuing calibration certicates based on the test method. In this paper we have presented a detailed procedure to evaluate the uncertainty in the calibration of the positional deviations of linear axes of numerically controlled machine tools. The method complies strictly with the GUMs recommendations: it is derived from a model for the measurand, whose uncertainty is obtained through application of the law of propagation of un-

ym y yM .

Careful consideration of (A.4) and (A.5) yields the pdf fz (z). It is of the form fz (z) = (wx wy )1 z2 ln(R), (A.6)

where R depends on z and on whether xm and xM are positive or negative (y is always positive). The function R is given in Table A.1. Unfortunately from the pdf fz (z) we cannot obtain an analytical expression for the standard uncertainty u[z]; it has to be obtained numerically. Let us assume that the calibration temperature is (18 0.5) C, such that T = x = 2 C, xm =

I. Lira, G. Cargill / Precision Engineering 28 (2004) 232239 Table A.1 Function R to be inserted in Eq. (A.6) R zxM yM /(1 z) yM /ym (1 z)/(zxm ym ) zxM yM /(1 z) xM /xm (1 z)/(zxm ym ) zxM yM /(1 z) yM /ym zxm yM /(1 z) (1 z)/(zxM ym ) yM /ym zxm yM /(1 z) (1 z)/(zxM ym ) xm /xM zxm yM /(1 z) Range for z (zMM , zMm ) (zMm , zmM ) (zmM , zmm ) (zMM , zmM ) (zmM , zMm ) (zMm , zmm ) (zMM , zMm ) (zMm , zmm ) (zmm , zmM ) (zMm , zMM ) (zMM , zmm ) (zmm , zmM ) (zMm , zmm ) (zmm , zMM ) (zMM , zmM ) Applies if xm yM < xM ym and 0 < xm

239

where V and E represent the variance and expectation operators, respectively. Now, one can assume that the pdf associated with is uniform, centered at = 0 and of width 2 max . This gives E(cos2 ) = 1 (2max ) cos2 d (A.8)

xM ym < xm yM and 0 < xm

=
xm < 0 < xM

(sin max cos max ) 1 + 2max 2 1 (2max ) sin max . max

and E(cos ) = cos d = (A.9)

xM yM < xm ym and xM < 0

xm ym < xM yM and xM < 0

In this table, zab is dened as 1/(1 + xa yb ), where subscripts a and b are both either m or M.

2.5 C and xM = 1.5 C. Assume also that = y = 11.7 106 C1 with w = 2.34 106 C1 , such that ym = 10.53 106 C1 and yM = 12.87 106 C1 . The standard deviation of fz (z) turns out to be u[ft ] = 3.64106 , the same result one gets from (15). If instead the temperature difference T is estimated to be zero with xm = 1 C, xM = 1 C and the same values for ym and yM , the standard deviation of fz (z) is u[ft ] = 6.77 106 , whereas from (15) one gets u[ft ] = 6.75 106 . Finally, if again xm = 1 C and xM = 1 C but the spread of is increased to 50% from its nominal value, u[ft ] increases to 7.03 106 while the value one obtains from (15) remains equal to 6.75 106 . Therefore, we conclude that in most cases the use of the LPU is appropriate for the NDE correction, unless one assigns a very large uncertainty to the coefcient . Consider now the cosine error. This is treated in paragraph F.2.4.4 of the GUM, which analysis is rather complicated. A simpler procedure, taken from Section 3.5 of Reference [5], consists of obtaining the uncertainty u[fa ] as u2 [fa ] = V(fa ) = E(cos2 ) [E(cos )]2 , (A.7)

To be consistent, the value of fa should then be taken not as 1, but as the expectation of cos . This gives Eq. (21). Substitution of (A.8) and (A.9) in (A.7) gives Eq. (20). Evidently, other formulas would apply if another pdf is used for the misalignment angle. We note nally that one might think of treating the cosine error with a uniform pdf. Indeed, it is possible to model the measurand as Lo = L ca , where L is the measured value and ca = L(1 cos ) is an additive alignment correction. One then gets u2 [Lo ] = u2 [L] + u2 [ca ]. A uniform pdf of width wa for ca gives u2 [ca ] = w2 a /12. This would be incorrect, however, because it would mean centering the pdf around the value zero for a quantity that is always positive. References
[1] ISO 230-2. Test code for machine toolsPart 2: determination of accuracy and repeatability of positioning of numerically controlled axes. Geneva: International Organization for Standardization; 1997. [2] ISO Guide to the Expression of Uncertainty in Measurement. Geneva: International Organization for Standardization; 1993 and 1995. [3] International Vocabulary of Basic and General Terms in Metrology. Geneva: International Organization for Standardization; 1993. [4] EA-4/02. Expression of the uncertainty of measurement in calibration. European Co-operation for Accreditation; 1999. [5] Lira I. Evaluating the uncertainty of measurement: fundamentals and practical guidance. Bristol: Institute of Physics Publishing; 2002.